In this thesis, we designed a neural network development environment(NNDE) based on the high-level neural network(NN) description language, MENTAL, and implemented a prototype. Though the MENTAL provides very high-level abstractions of target NNs, previous implementations were very limited and only worked on a few platforms. Our NNDE consists of a translator translating the NN descriptions into the C programs, and a run-time environment. We added output buffer mechanisms to basic MENTAL virtual machine to simulate time-delayed neural connections, and then automated the visualization of the NN by exploiting the NN's shape description written in MENTAL.